为什么 model.predict() 的精度比具有相同 x_train 的 model.fit() 更差？

Question

images receives an array of images loaded with cv2

images=np.array(images)
labels=np.array(labels)

idLabels=[]
for i in labels:
    idLabels.append(dicTipos[i])
labels=np.array(idLabels)

images = np.array(images, dtype = 'float32')
print("images done")
print("labels", labels)
labels = np.array(labels, dtype = 'int32')


x_train=images
y_train=labels

I defined the model and after that I use fit()

model.compile(optimizer='adam',
              loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=False),
              metrics=['accuracy'])


history = model.fit(x_train, y_train, epochs=10)

The final line of the output is:

Epoch 10/10
25/25 [==============================] - 101s 4s/step - loss: 0.0047 - accuracy: 0.9987

Immediately I use predict() function and the accurary is really bad:

predicted=model.predict(x_train)
rounded_predictions=np.argmax(predicted,axis=1)
temp = sum(y_train == rounded_predictions)
temp=temp/len(y_train)
print("Accuracy:  ", temp)

Output:

Accuracy:   0.12625

I don't know why this (worse accuracy than fit) could happen if I'm setting the same x_train for training and the same x_train for testing

Answer 1

You should directly use the accuracy metrics that you compiled with the model. This will ensure you get accuracy evaluated the same way when training and testing

Try:

perf = model.evaluate(x_train, y_train,return_dict = True)
print (perf)